How do scopes sample at tens of GHz?

It's been a while since I looked at high speed scopes. But last I heard, they were topping out around 5Gsamples/sec for a very expensive tek. Now I see ads in microwave magazines for scopes that claim sampling rates of up to 100GHz. I don't know anyone who makes an A/D chip that will sample at anything close to 100GHz or a memory manufacturer who can handle that kind of data rate. So what's the trick? Are they ping-ponging a large bank of slower A/D's ? Does the enormous sample rate only apply to repetitive waveforms? Or can I really put a single 100psec pulse into these things and get 10 clean points on it?

No. I think it is a repetitive rate. Thus, for example it may sample 1000 samples per sweep, but each sample is delayed by 10ps from the previous sweep. Thus over a series of sweeps you get the equivalent of a 100GHz sample rate. But the aperture (sample) period must still be very short (likely less than 10ps) to get a clean sample and that's obviously not a trivial requirement.

But that's usually not a serious limitation since it's not often that you have only a single 100ps pulse that you need to see.